Before getting too far into the actual review, we thought it appropriate to give a short and succinct overview on the technical specs and new features of the NVIDIA GeForce3.

The GeForce3 is produced using the 0.15 micron process (compared to the 0.18 micron design of the GeForce2), and features an astounding 57 million transistors. In terms of physical design, the GeForce3 is extremely similar to the previous GeForce2. Both chips have quad pixel pipelines and support high-speed DDR memory. Somewhat surprisingly, the GeForce3 ships at a clock speed of 200 MHz, a full 50 MHz slower than the high-end GeForce2 ULTRA and on par with both the GeForce2 Pro and GTS models. The GeForce3's 460 MHz memory speed is also right on par with the GeForce2 ULTRA. These speeds may seem to put the GeForce3 at a bit of a performance disadvantage, but the true difference between the GeForce2 and GeForce3 is not pure horsepower but rather technological enhancements that have the potential to increase overall gaming performance.

Possibly the most important new feature of the GeForce3 is its nfiniteFX Engine, which includes Programmable Vertex Shaders and Pixel Shaders as its highlights. Both of these new features allow developers to make use programmable functions to allow more realistic character movement and animation. The Vertex Shader means a higher and more realistic level of vertex melding (such as the joints in a 3D character) while the Pixel function allows for a greater number of shaded pixels to be processed and also supports specialized effects through its programmability.

In addition to the above technology, NVIDIA has also included a few hardware improvements over previous GeForce2 products. Memory bandwidth is a common limitation with higher-end GeForce2 boards, and the new Lightspeed Memory Architecture is used to combat this with the GeForce3. This architecture is comprised of a few important components, including a Crossbar Memory Controller, Higher Order Surfaces, Lossless Z Compression, and Z-Occlusion Culling.

Possibly the most important of them all, especially for current 3D games, is the Crossbar Memory Controller. In order to alleviate the strain on the card's memory bus, even at ultra-high 460 MHz speeds, NVIDIA has implemented four independent memory controllers, which share the data load and allows each one to process smaller pieces of data. This can yield a higher rate of efficiency, as the controllers share the total memory bandwidth and while one is processing data the others are freed up for additional duties. This system is very similar to the memory controller found in the upcoming nForce chipset.

Both Lossless Z Compression and Z-Occlusion Culling are forms of minimizing potential memory bandwidth requirements. Lossless Z Compression is basically a method of compressing Z-Buffer memory transfers (up to 4:1) and freeing up bandwidth for other uses. Z-Occlusion Culling is basically another name for Hidden Surface Removal or Overdraw, or similar technologies using a myriad selection of terms. This technology removes pixels that are not displayed on the screen (or "occluded") and save both memory bandwidth requirements and GPU processing demands.

Higher Order Surfaces is a feature with future potential for programmers, but don't expect it to yield any real results in current games. This essentially boils down to the GeForce3 being able to move away from standard triangle setup, and instead accept "higher order" formulas to display geometric objects.

High-Resolution Antialiasing (or HRAA for short) is another feature that has been improved with the GeForce3. Instead of high jacking the GeForce2's GPU for HRAA usage, the GeForce3 has instituted hardware support directly in the chip itself. The wider data paths of the GeForce3 can easily accommodate the extra pixels generated by HRAA and a new Quincunx AA pattern has been hard-coded into the GeForce3 architecture. This Quincunx AA cuts a swath between 2X and 4X HRAA image quality, while offering a minimal performance hit. In our testing, Quincunx AA seems to offer an excellent tradeoff between performance and image quality.

Many of the GeForce3's technological improvements hinge greatly on DirectX 8 support. In fact, both the supported features of DirectX 8 and those of the GeForce3 look eerily similar. The only major change for DirectX 8 seems to be the emphasis on the programmability of Vertex and Pixel Shaders, as opposed to the static or fixed function models used in previous versions. This close relationship between the NVIDIA GeForce3 and the Microsoft DirectX 8 API means that for the foreseeable future at least, the GeForce3 will be the only truly DirectX 8-compliant video card.